Ungoogled Chromium127编译指南 Windows篇 - 链接工具修复（十三）

标签：Chromium127 resource undname Windows Ungoogled pdbutil exe path line

1. 引言

在完成Rust编译器包装器的修复后，我们遇到了编译过程最后阶段的链接问题。在Windows环境下，链接器无法正确找到llvm-pdbutil.exe和undname.exe这两个重要工具。本文将详细介绍如何解决这个问题。

2. 问题分析

2.1 错误表现

链接阶段报错找不到必要的工具：

llvm-pdbutil.exe无法被正确定位
undname.exe工具无法被找到和使用

FAILEC: gen/chrome/chrome_resource_allowlist.txt
C:/Users/ym/AppData/Local/Programs/Python/Python39/python.exe ../../tools/resources/generate_resource_allowlist.py -o gen/chrome/chrome_resource_allowlist.txt chrome.dll.pdb
Traceback (most recent call last):
  File "C:\ungoogled-chromium-windows\build\src\tools\resources\generate_resource_allowlist.py", line 165, in <module>
    main()
  File "C:\ungoogled-chromium-windows\build\src\tools\resources\generate_resource_allowlist.py", line 161, in main
    WriteResourceAllowlist(args)
  File "C:\ungoogled-chromium-windows\build\src\tools\resources\generate_resource_allowlist.py", line 140, in WriteResourceAllowlist
    resource_ids.update(func(input))
  File "C:\ungoogled-chromium-windows\build\src\tools\resources\generate_resource_allowlist.py", line 61, in GetResourceAllowlistPDB
    pdbutil = subprocess.Popen(
  File "C:\Users\ym\AppData\Local\Programs\Python\Python39\lib\subprocess.py", line 951, in __init__
    self._execute_child(args, executable, preexec_fn, close_fds,
  File "C:\Users\ym\AppData\Local\Programs\Python\Python39\lib\subprocess.py", line 1420, in _execute_child
    hp, ht, pid, tid = _winapi.CreateProcess(executable, args,
OSError: [WinError 193] %1 不是有效的 Win32 应用程序。
ninja: build stoppeC: subcommand failed.

C:\ungoogled-chromium-windows\build\src>exit
Traceback (most recent call last):
  File "C:\ungoogled-chromium-windows\build.py", line 464, in <module>
    main()
  File "C:\ungoogled-chromium-windows\build.py", line 455, in main
    _run_build_process('third_party\\ninja\\ninja.exe', '-C', 'out\\Default', 'chrome',
  File "C:\ungoogled-chromium-windows\build.py", line 78, in _run_build_process
    subprocess.run(('cmd.exe', '/k'),
  File "C:\Users\ym\AppData\Local\Programs\Python\Python39\lib\subprocess.py", line 528, in run
    raise CalledProcessError(retcode, process.args,
subprocess.CalledProcessError: Command '('cmd.exe', '/k')' returned non-zero exit status 1.

2.2 错误原因

工具路径解析逻辑未考虑操作系统差异
资源生成脚本中的工具调用方式不正确

3. 修复方案

3.1 修改generate_resource_allowlist.py

需要修改资源生成脚本中的工具调用逻辑：

# 修改前
pdbutil = subprocess.Popen(['llvm-pdbutil', 'dump', '-publics', path],
                         stdout=subprocess.PIPE)

# 修改后
pdbutil_exe = 'llvm-pdbutil.exe' if sys.platform == 'win32' else 'llvm-pdbutil'
pdbutil_path = os.path.join(llvm_bindir, pdbutil_exe)
pdbutil = subprocess.Popen([pdbutil_path, 'dump', '-publics', path],
                         stdout=subprocess.PIPE)

3.2 完善工具调用

对undname工具的调用也需要类似修改：

修改前

  for name in names:
    undname = subprocess.Popen(['undname', name],
                               stdout=subprocess.PIPE)

修改后

  undname_exe = 'undname.exe' if sys.platform == 'win32' else 'undname'
  for name in names:
    undname = subprocess.Popen([undname_exe, name],
                             stdout=subprocess.PIPE)

3.3 完整脚本

#!/usr/bin/env python
# Copyright 2016 The Chromium Authors
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.

__doc__ = """generate_resource_allowlist.py [-o OUTPUT] INPUTS...

INPUTS are paths to unstripped binaries or PDBs containing references to
resources in their debug info.

This script generates a resource allowlist by reading debug info from
INPUTS and writes it to OUTPUT.
"""

# Allowlisted resources are identified by searching the input file for
# instantiations of the special function ui::AllowlistedResource (see
# ui/base/resource/allowlist.h).

import argparse
import os
import subprocess
import sys

import ar

llvm_bindir = os.path.join(os.path.dirname(sys.argv[0]), '..', '..',
                           'third_party', 'llvm-build', 'Release+Asserts',
                           'bin')


def ExtractAllowlistFromFile(path, resource_ids):
  with open(path, 'rb') as f:
    data = f.read()
  # When symbol_level=0, only mangled names exist.
  # E.g.: _ZN2ui19AllowlistedResourceILi22870EEEvv
  prefix = b'AllowlistedResourceILi'
  start_idx = 0
  while start_idx != -1:
    start_idx = data.find(prefix, start_idx)
    if start_idx != -1:
      end_idx = data.find(b'E', start_idx)
      resource_ids.add(int(data[start_idx + len(prefix):end_idx]))
      start_idx = end_idx


def GetResourceAllowlistELF(path):
  # Produce a resource allowlist by searching for debug info referring to
  # AllowlistedResource.
  # This used to use "readelf -p .debug_str", but it doesn't seem to work with
  # use_debug_fission=true. Reading the raw file is faster anyways.
  resource_ids = set()
  ExtractAllowlistFromFile(path, resource_ids)
  return resource_ids


def GetResourceAllowlistPDB(path):
  # Produce a resource allowlist by using llvm-pdbutil to read a PDB file's
  # publics stream, which is essentially a symbol table, and searching for
  # instantiations of AllowlistedResource. Any such instantiations are demangled
  # to extract the resource identifier.
  # 在 Windows 环境下添加 .exe 扩展名
  pdbutil_exe = 'llvm-pdbutil.exe' if sys.platform == 'win32' else 'llvm-pdbutil'
  pdbutil_path = os.path.join(llvm_bindir, pdbutil_exe)
  pdbutil = subprocess.Popen(
        [pdbutil_path, 'dump', '-publics', path],
        stdout=subprocess.PIPE)

  names = set()
  for line in pdbutil.stdout:
    line = line.decode('utf8')
    # Read a line of the form
    # "733352 | S_PUB32 [size = 56] `??$AllowlistedResource@$0BFGM@@ui@@YAXXZ`".
    if '`' not in line:
      continue
    sym_name = line[line.find('`') + 1:line.rfind('`')]
    # Under certain conditions such as the GN arg `use_clang_coverage = true` it
    # is possible for the compiler to emit additional symbols that do not match
    # the standard mangled-name format.
    # Example: __profd_??$AllowlistedResource@$0BGPH@@ui@@YAXXZ
    # C++ mangled names are supposed to begin with `?`, so check for that.
    if 'AllowlistedResource' in sym_name and sym_name.startswith('?'):
      names.add(sym_name)
  exit_code = pdbutil.wait()
  if exit_code != 0:
    raise Exception('llvm-pdbutil exited with exit code %d' % exit_code)

  resource_ids = set()
  undname_exe = 'undname.exe' if sys.platform == 'win32' else 'undname'
  for name in names:
    undname = subprocess.Popen([undname_exe, name],
                             stdout=subprocess.PIPE)
    found = False
    for line in undname.stdout:
      line = line.decode('utf8')
      # Read a line of the form
      # "void __cdecl ui::AllowlistedResource<5484>(void)".
      prefix = ' ui::AllowlistedResource<'
      pos = line.find(prefix)
      if pos == -1:
        continue
      try:
        resource_ids.add(int(line[pos + len(prefix):line.rfind('>')]))
      except ValueError:
        continue
      found = True
      break
    exit_code = undname.wait()
    if exit_code != 0:
      raise Exception('llvm-undname exited with exit code %d' % exit_code)
    if not found:
      raise Exception('Unexpected undname output')
  return resource_ids


def GetResourceAllowlistFileList(file_list_path):
  # Creates a list of resources given the list of linker input files.
  # Simply grep's them for AllowlistedResource<...>.
  with open(file_list_path) as f:
    paths = f.read().splitlines()

  paths = ar.ExpandThinArchives(paths)

  resource_ids = set()
  for p in paths:
    ExtractAllowlistFromFile(p, resource_ids)
  return resource_ids


def WriteResourceAllowlist(args):
  resource_ids = set()
  for input in args.inputs:
    with open(input, 'rb') as f:
      magic = f.read(4)
      chunk = f.read(60)
    if magic == b'\x7fELF':
      func = GetResourceAllowlistELF
    elif magic == b'Micr':
      func = GetResourceAllowlistPDB
    elif magic == b'obj/' or b'/obj/' in chunk:
      # For secondary toolchain, path will look like android_clang_arm/obj/...
      func = GetResourceAllowlistFileList
    else:
      raise Exception('unknown file format')

    resource_ids.update(func(input))

  # The last time this broke, exactly two resources were still being found.
  if len(resource_ids) < 100:
    raise Exception('Suspiciously few resources found. Likely an issue with '
                    'the regular expression in this script. Found: ' +
                    ','.join(sorted(resource_ids)))
  for id in sorted(resource_ids):
    args.output.write(str(id) + '\n')


def main():
  parser = argparse.ArgumentParser(usage=__doc__)
  parser.add_argument('inputs', nargs='+', help='An unstripped binary or PDB.')
  parser.add_argument('-o',
                      dest='output',
                      type=argparse.FileType('w'),
                      default=sys.stdout,
                      help='The resource list path to write (default stdout)')

  args = parser.parse_args()
  WriteResourceAllowlist(args)


if __name__ == '__main__':
  main()

4. 验证修复

4.1 测试步骤

修改generate_resource_allowlist.py文件
确保工具存在于正确路径
重新执行编译命令：

python build.py --tarball

4.2 验证要点

确认llvm-pdbutil.exe能被正确调用
验证undname.exe工具运行正常
检查资源生成过程完整性

5. 可能遇到的问题

5.1 工具路径问题

确保LLVM工具链安装完整
验证工具所在目录已添加到PATH
检查文件访问权限设置

5.2 环境配置问题

确认Python环境变量设置
验证Windows系统兼容性
检查构建脚本权限

6. 结语

通过本文的指导，我们成功解决了Ungoogled Chromium编译过程中最后链接阶段的工具调用问题。这个修复确保了编译过程能够顺利完成。

至此，我们已经解决了Chromium编译过程中的主要技术难点。在下一篇文章中，我们将验证编译完成的Ungoogled Chromium，探索其特性。请确保按本文的步骤正确修复链接工具的问题，为最终的编译成功奠定基础。

标签：Chromium127,resource,undname,Windows,Ungoogled,pdbutil,exe,path,line
From： https://blog.csdn.net/qqyy_sj/article/details/144441455